FAQ

#Does Agentower see our code or prompts?

Not your code, and not your prompts unless you choose to. By default it gets status, usage and counts of what each agent did — see What Agentower sends and keeps. Two settings send more, and both are off until someone switches them on:

• Agent chats (workspace-wide, owners only) — the agent's questions and their options, its final message each turn (for the chat), what an approval is for, and a plan waiting for approval.
• Work details (per project, owners and that project's approvers) — file paths, commit messages and failed test names, plus the first 600 characters of the instruction that started each turn and of the agent's closing summary, with anything that looks like a secret replaced by [hidden] on your computer before it's sent. Only people who may approve the project see them. See The work record.

File contents and diffs are never sent, at any setting. (With Agent chats on, the agent's final message can quote code it shows you, and an approval shows the command it's for.)

#Does it replace Claude Code, Codex or Cursor?

No. Agentower doesn't write code, run terminals or run your agents. Your developers keep their tools; Agentower shows which of their agents needs someone, and lets the right person answer.

#If Agentower is slow or down, do our agents get stuck?

No. The hook gives up after a few seconds and the agent carries on as it would without Agentower, asking in the terminal. Approvals and questions can be answered in Agentower for up to 60 minutes (9 for Codex or an older hook); after that, answer in the terminal.

#What if someone answers in the terminal too?

That's fine: whichever answer comes first wins, and the other is closed. See Answer agents from Agentower.

#Can I answer an agent's questions or approve its plan from Agentower?

Yes, with Agent chats on and the agent's computer allowing answers from Agentower. When Claude asks a multiple-choice question, its chat shows the options: pick one (or several, where it allows), or type your own answer. When it has a plan ready, the chat shows the plan with Approve plan and Keep planning. Either way it's as if you'd answered in the terminal, and the first answer wins. A plan over 12,000 characters, or one with characters Agentower had to remove, can only be refused from the app — approve it in the terminal. See Answer its questions and Approve a plan.

#Who can approve or message an agent?

Only owners and approvers, only when Agent chats is on for the workspace, and — if the project names its approvers — only those people and owners. It needs a sign-in from the last 7 days. See Roles and approvers.

The agent's computer has the last word: unless whoever runs it allows answers from Agentower there (own for just themselves, team for the project's approvers too), nothing from the app reaches it. It's off until they choose. See Each computer decides too.

Allow and Deny — and answers to its questions and plans — answer the agent's prompt just as the terminal would. A message is different: the agent treats it as a request, not as your consent, so for some actions it may still ask in the terminal — see A message is a request, not a yes.

#Can a message reach an agent that's sitting idle?

Yes, if its computer allows answers from Agentower. If the agent finished its turn while Agent chats was on, it listens for a message from the app for up to 4 hours. For delivery straight into the session at any time, start Claude Code with the optional Agentower channel (Claude Code Channels, research preview).

#An agent is finished, but the board still shows it as Working. What can I do?

Owners and the project's approvers can press Move to Idle in its details panel (or Move all to Idle on the Starting and Failed headers), or Move to Inactive to set it aside. Nothing stops on the computer: its next real step moves it on as usual. If it keeps happening, the session may have closed without telling Agentower — it closes by itself after a while with no signal; see When an agent closes by itself.

#Does every developer need to change how they work?

No. Each person connects their machine once. After that they use their tools as before, and can answer in the terminal or in Agentower, whichever is closer.

#Can I use it on my phone?

Yes. Agentower runs in the browser, and the board, the details panel and answering all work on a phone. There's no app to install.

#Which tools work?

Claude Code, Codex and Cursor, with different levels of support — see Connecting tools. Anything else can report through the REST API or the MCP server.

#Do you charge for tokens?

No. Your agents run on the plans and keys you already pay for; Agentower adds no fee on tokens. Reading and acting through Agentower's own API is priced per call — see API pricing. The money figures in Agentower are "API value" — what the same tokens would cost on the pay-as-you-go API — not what you pay on a plan.

#What does it cost?

Signing up gives no access until you add a card in Stripe Checkout. That starts Solo with a free period: a 14-day free trial or, during the beta while one of the 5 free places is open, 6 months free. Nothing is charged during the free period. When it ends, your card is charged $5 a month + VAT for Solo, and every month after, until you cancel. Cancel any time before the free period ends (Settings → Billing → Manage billing) and nothing is charged; reminder emails go out 3 days and again 24 hours before it ends, each saying how to cancel and how to delete everything. An email address gets one free period only. During the 14-day trial, access is one day at a time: each day you press Continue for 24 hours (a 24-hour pass, up to the end of the trial), and between passes the workspace is paused and your agents' reports are not taken. The 6-month free places have no daily passes. Payments run through Stripe Managed Payments: Stripe is the seller of record, adds VAT at checkout at your country's rate and issues the receipts. Solo is for one person. Team is $12 per seat a month + VAT, billed monthly with no minimum and no free period — a second person needs Team. Enterprise is $20 per seat a month + VAT, billed annually ($240 a seat a year + VAT), for at least 10 seats, with no free period, and an owner can buy it in Settings → Billing. API calls beyond the included amount are priced per call, with volume discounts; reporting your agents and public stats are always free. See Plans and billing and API pricing. If the free period ends without a paid plan, the workspace pauses: reports stop and the board asks you to choose a plan. Solo, Team and Enterprise are paid by card at checkout, with a full refund within 14 days of your first payment (once per workspace and per person).

#Can I get my data out?

Yes, as CSV or JSON — agents, their history and work, usage, projects, the team, guardrails, channels, routines, computers, keys, plan limits and what Agentower keeps about you. Not exported: agent chats (never — they're held only while the agent is open) and routines' briefs (read them in the app). See Export your data. You can also delete your account yourself.

#Can a bot or script answer agents?

Yes, with an agent-actions key made by an owner or approver. It can do only what that person could do in the app, expires after 30 days, and every action is in the audit log. See Agent actions API.

#Can I use Agentower without the hook, just to read stats?

Yes. A public stats key reads only the counters you choose to publish, for example to show "agents working right now" on your own website — see Public stats.

#Where is our data stored?

With Cloudflare, in the EU.